Hyderabad: ByteDance, the parent company of TikTok, has released OmniHuman-1-- an AI system that can generate lifelike videos of people doing activities, such as singing, dancing, talking, gesturing, and more from a single photo.
On February 3, 2025, a research paper was submitted by the company's researchers to Cornell University. It mentioned that OmniHuman proves to be a better AI video generator compared to others as it "scales up data by mixing motion-related conditions into the training phase."
Based on the weak signal inputs, the AI video generator produces an extremely realistic human video, including audio. Moreover, this AI video generator supports image inputs of various aspect ratios that include portraits, half-body, or full-body images, showcasing a lifelike video in high-quality resolutions.
The researchers have published video samples and a flowchart on GitHub, showcasing the working of the video generator. The page displays various video samples generated by OmniHuman-1, which include portrait videos, people talking, and more.
Albert Einstein's AI-generated video, that depicts about human emotions using accurate human gestures. (Credit: GitHub/OmniHuman Lab) For instance, an AI-generated video of Albert Einstein under the 'talking category', depicts how he stands in front of a blackboard and talks about the importance of human emotions. In the video, it is clearly visible that Einstein is punctuating his words using hand gestures and subtle facial expressions making it extremely realistic.
The OmniHuman is expected to be a great threat to other AI-video generators, such as Veo by Google, Sora by OpenAI, Runway, Descript, and more. With the launch of OmniHuman-1, the future of AI-video generators seems to be very bright producing extremely realistic human-like videos, including audio and human gestures.
Also Read:
